What Supercomputer means

Supercomputer is the hosted execution plane for TARX.

Use it when:

local hardware is not enough
the workload needs more reasoning depth
the job is longer-running or more multimodal
a team wants shared hosted capacity

The important product truth is simple:

Supercomputer uses the same familiar request shape as local TARX where the route supports it.

Hosted Supercomputer is measured in Joules. Familiar indicators such as tokens, seconds, images, jobs, or video duration may appear for comparison, but Joules are the native hosted compute unit.

What developers should expect

Developers should not need to learn TARX’s internal topology to use Supercomputer.

They should expect:

one familiar request shape
hosted execution through https://tarx.com/api/v1
TARX API keys for hosted access
the same app moving between local and hosted without rewrite

What enterprise should expect

Enterprise should be able to use Supercomputer in two ways:

TARX-managed hosted compute
Enterprise BYO agreements

That second mode matters because many companies already have:

provider contracts
cloud commitments
region constraints
internal model approvals

TARX should act as the control, policy, and evidence layer across those agreements instead of forcing a rip-and-replace.

Local versus Supercomputer

Local TARX

best for first-run development
no hosted key required
local-first files, context, and private state

Supercomputer

best for heavier reasoning and shared capacity
hosted API key required
same request shape where the route supports it

What belongs in the request surface

At minimum:

GET /health
POST /v1/chat/completions

Over time, the same runtime model should extend to:

embeddings
voice
vision
image workflows
video jobs
sessions
memory
files
tasks
tools
route evidence

Pricing posture

Supercomputer should be positioned as optional power, not as a generic hosted-token clone.

The clean public shape is:

TARX Free starts on the user’s computer
Supercomputer is permissioned power when the Computer is not enough
users can contribute compute to earn access
Power Bundles cover usage beyond allowance or contributed compute
Joules are the TARX-native hosted compute meter
familiar workload indicators belong in account, billing, and developer detail views as comparison detail
Enterprise Runtime is custom
enterprise BYO keeps the same request shape with customer-approved providers

Do not publish fake usage conversions. If a conversion is not settled, keep marketing language centered on optional power and use technical meters only where they help account or developer workflows.

Important truth boundary

Supercomputer should not be marketed as “everything always stays local.”

The truthful posture is:

local-first where supported
hosted when selected or required
enterprise-private when deployed accordingly